xref: /illumos-gate/usr/src/uts/i86pc/os/mp_machdep.c (revision da6c28aaf62fa55f0fdb8004aa40f88f23bf53f0)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #define	PSMI_1_6
29 #include <sys/smp_impldefs.h>
30 #include <sys/psm.h>
31 #include <sys/psm_modctl.h>
32 #include <sys/pit.h>
33 #include <sys/cmn_err.h>
34 #include <sys/strlog.h>
35 #include <sys/clock.h>
36 #include <sys/debug.h>
37 #include <sys/rtc.h>
38 #include <sys/x86_archext.h>
39 #include <sys/cpupart.h>
40 #include <sys/cpuvar.h>
41 #include <sys/cmt.h>
42 #include <sys/cpu.h>
43 #include <sys/disp.h>
44 #include <sys/archsystm.h>
45 #include <sys/machsystm.h>
46 #include <sys/sysmacros.h>
47 #include <sys/memlist.h>
48 #include <sys/param.h>
49 #include <sys/promif.h>
50 #if defined(__xpv)
51 #include <sys/hypervisor.h>
52 #endif
53 #include <sys/mach_intr.h>
54 #include <vm/hat_i86.h>
55 #include <sys/kdi_machimpl.h>
56 
57 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
58 
59 /*
60  *	Local function prototypes
61  */
62 static int mp_disable_intr(processorid_t cpun);
63 static void mp_enable_intr(processorid_t cpun);
64 static void mach_init();
65 static void mach_picinit();
66 static int machhztomhz(uint64_t cpu_freq_hz);
67 static uint64_t mach_getcpufreq(void);
68 static void mach_fixcpufreq(void);
69 static int mach_clkinit(int, int *);
70 static void mach_smpinit(void);
71 static int mach_softlvl_to_vect(int ipl);
72 static void mach_get_platform(int owner);
73 static void mach_construct_info();
74 static int mach_translate_irq(dev_info_t *dip, int irqno);
75 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
76     psm_intr_op_t, int *);
77 static void mach_notify_error(int level, char *errmsg);
78 static hrtime_t dummy_hrtime(void);
79 static void dummy_scalehrtime(hrtime_t *);
80 static void cpu_idle(void);
81 static void cpu_wakeup(cpu_t *, int);
82 #ifndef __xpv
83 static void cpu_idle_mwait(void);
84 static void cpu_wakeup_mwait(cpu_t *, int);
85 #endif
86 /*
87  *	External reference functions
88  */
89 extern void return_instr();
90 extern uint64_t freq_tsc(uint32_t *);
91 #if defined(__i386)
92 extern uint64_t freq_notsc(uint32_t *);
93 #endif
94 extern void pc_gethrestime(timestruc_t *);
95 extern int cpuid_get_coreid(cpu_t *);
96 extern int cpuid_get_chipid(cpu_t *);
97 
98 /*
99  *	PSM functions initialization
100  */
101 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
102 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
103 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
104 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
105 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
106 void (*psminitf)()		= mach_init;
107 void (*picinitf)() 		= return_instr;
108 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
109 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
110 void (*send_dirintf)() 		= return_instr;
111 void (*setspl)(int)		= (void (*)(int))return_instr;
112 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
113 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
114 void (*kdisetsoftint)(int, struct av_softinfo *)=
115 	(void (*)(int, struct av_softinfo *))return_instr;
116 void (*setsoftint)(int, struct av_softinfo *)=
117 	(void (*)(int, struct av_softinfo *))return_instr;
118 int (*slvltovect)(int)		= (int (*)(int))return_instr;
119 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
120 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
121 int (*psm_disable_intr)(int)	= mp_disable_intr;
122 void (*psm_enable_intr)(int)	= mp_enable_intr;
123 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
124 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
125 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
126 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
127 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
128 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
129 int (*psm_get_clockirq)(int) = NULL;
130 int (*psm_get_ipivect)(int, int) = NULL;
131 
132 int (*psm_clkinit)(int) = NULL;
133 void (*psm_timer_reprogram)(hrtime_t) = NULL;
134 void (*psm_timer_enable)(void) = NULL;
135 void (*psm_timer_disable)(void) = NULL;
136 void (*psm_post_cyclic_setup)(void *arg) = NULL;
137 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
138     int *) = mach_intr_ops;
139 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
140     return_instr;
141 
142 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
143 void (*hrtime_tick)(void)	= return_instr;
144 
145 /*
146  * True if the generic TSC code is our source of hrtime, rather than whatever
147  * the PSM can provide.
148  */
149 #ifdef __xpv
150 int tsc_gethrtime_enable = 0;
151 #else
152 int tsc_gethrtime_enable = 1;
153 #endif
154 int tsc_gethrtime_initted = 0;
155 
156 /*
157  * True if the hrtime implementation is "hires"; namely, better than microdata.
158  */
159 int gethrtime_hires = 0;
160 
161 /*
162  * Local Static Data
163  */
164 static struct psm_ops mach_ops;
165 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
166 static ushort_t mach_ver[4] = {0, 0, 0, 0};
167 
168 /*
169  * If non-zero, idle cpus will become "halted" when there's
170  * no work to do.
171  */
172 int	idle_cpu_use_hlt = 1;
173 
174 #ifndef __xpv
175 /*
176  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
177  */
178 int	idle_cpu_prefer_mwait = 1;
179 #endif
180 
181 /*ARGSUSED*/
182 int
183 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
184 {
185 	switch (hw) {
186 	case PGHW_IPIPE:
187 		if (x86_feature & (X86_HTT)) {
188 			/*
189 			 * Hyper-threading is SMT
190 			 */
191 			return (1);
192 		} else {
193 			return (0);
194 		}
195 	case PGHW_CHIP:
196 		if (x86_feature & (X86_CMP|X86_HTT))
197 			return (1);
198 		else
199 			return (0);
200 	case PGHW_CACHE:
201 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
202 			return (1);
203 		else
204 			return (0);
205 	default:
206 		return (0);
207 	}
208 }
209 
210 /*
211  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
212  * If pghw_type_t is an unsupported hardware type, then return -1
213  */
214 int
215 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
216 {
217 	id_t pgp_a, pgp_b;
218 
219 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
220 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
221 
222 	if (pgp_a == -1 || pgp_b == -1)
223 		return (-1);
224 
225 	return (pgp_a == pgp_b);
226 }
227 
228 /*
229  * Return a physical instance identifier for known hardware sharing
230  * relationships
231  */
232 id_t
233 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
234 {
235 	switch (hw) {
236 	case PGHW_IPIPE:
237 		return (cpuid_get_coreid(cpu));
238 	case PGHW_CACHE:
239 		return (cpuid_get_last_lvl_cacheid(cpu));
240 	case PGHW_CHIP:
241 		return (cpuid_get_chipid(cpu));
242 	default:
243 		return (-1);
244 	}
245 }
246 
247 int
248 pg_plat_hw_level(pghw_type_t hw)
249 {
250 	int i;
251 	static pghw_type_t hw_hier[] = {
252 		PGHW_IPIPE,
253 		PGHW_CACHE,
254 		PGHW_CHIP,
255 		PGHW_NUM_COMPONENTS
256 	};
257 
258 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
259 		if (hw_hier[i] == hw)
260 			return (i);
261 	}
262 	return (-1);
263 }
264 
265 /*
266  * Return 1 if CMT load balancing policies should be
267  * implemented across instances of the specified hardware
268  * sharing relationship.
269  */
270 int
271 pg_plat_cmt_load_bal_hw(pghw_type_t hw)
272 {
273 	if (hw == PGHW_IPIPE ||
274 	    hw == PGHW_FPU ||
275 	    hw == PGHW_CHIP ||
276 	    hw == PGHW_CACHE)
277 		return (1);
278 	else
279 		return (0);
280 }
281 
282 
283 /*
284  * Return 1 if thread affinity polices should be implemented
285  * for instances of the specifed hardware sharing relationship.
286  */
287 int
288 pg_plat_cmt_affinity_hw(pghw_type_t hw)
289 {
290 	if (hw == PGHW_CACHE)
291 		return (1);
292 	else
293 		return (0);
294 }
295 
296 id_t
297 pg_plat_get_core_id(cpu_t *cpu)
298 {
299 	return ((id_t)cpuid_get_coreid(cpu));
300 }
301 
302 void
303 cmp_set_nosteal_interval(void)
304 {
305 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
306 	nosteal_nsec = 100000UL;
307 }
308 
309 /*
310  * Routine to ensure initial callers to hrtime gets 0 as return
311  */
312 static hrtime_t
313 dummy_hrtime(void)
314 {
315 	return (0);
316 }
317 
318 /* ARGSUSED */
319 static void
320 dummy_scalehrtime(hrtime_t *ticks)
321 {}
322 
323 /*
324  * Idle the present CPU until awoken via an interrupt
325  */
326 static void
327 cpu_idle(void)
328 {
329 	cpu_t		*cpup = CPU;
330 	processorid_t	cpun = cpup->cpu_id;
331 	cpupart_t	*cp = cpup->cpu_part;
332 	int		hset_update = 1;
333 
334 	/*
335 	 * If this CPU is online, and there's multiple CPUs
336 	 * in the system, then we should notate our halting
337 	 * by adding ourselves to the partition's halted CPU
338 	 * bitmap. This allows other CPUs to find/awaken us when
339 	 * work becomes available.
340 	 */
341 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
342 		hset_update = 0;
343 
344 	/*
345 	 * Add ourselves to the partition's halted CPUs bitmask
346 	 * and set our HALTED flag, if necessary.
347 	 *
348 	 * When a thread becomes runnable, it is placed on the queue
349 	 * and then the halted cpuset is checked to determine who
350 	 * (if anyone) should be awoken. We therefore need to first
351 	 * add ourselves to the halted cpuset, and and then check if there
352 	 * is any work available.
353 	 *
354 	 * Note that memory barriers after updating the HALTED flag
355 	 * are not necessary since an atomic operation (updating the bitmap)
356 	 * immediately follows. On x86 the atomic operation acts as a
357 	 * memory barrier for the update of cpu_disp_flags.
358 	 */
359 	if (hset_update) {
360 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
361 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
362 	}
363 
364 	/*
365 	 * Check to make sure there's really nothing to do.
366 	 * Work destined for this CPU may become available after
367 	 * this check. We'll be notified through the clearing of our
368 	 * bit in the halted CPU bitmask, and a poke.
369 	 */
370 	if (disp_anywork()) {
371 		if (hset_update) {
372 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
373 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
374 		}
375 		return;
376 	}
377 
378 	/*
379 	 * We're on our way to being halted.
380 	 *
381 	 * Disable interrupts now, so that we'll awaken immediately
382 	 * after halting if someone tries to poke us between now and
383 	 * the time we actually halt.
384 	 *
385 	 * We check for the presence of our bit after disabling interrupts.
386 	 * If it's cleared, we'll return. If the bit is cleared after
387 	 * we check then the poke will pop us out of the halted state.
388 	 *
389 	 * This means that the ordering of the poke and the clearing
390 	 * of the bit by cpu_wakeup is important.
391 	 * cpu_wakeup() must clear, then poke.
392 	 * cpu_idle() must disable interrupts, then check for the bit.
393 	 */
394 	cli();
395 
396 	if (hset_update && !CPU_IN_SET(cp->cp_mach->mc_haltset, cpun)) {
397 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
398 		sti();
399 		return;
400 	}
401 
402 	/*
403 	 * The check for anything locally runnable is here for performance
404 	 * and isn't needed for correctness. disp_nrunnable ought to be
405 	 * in our cache still, so it's inexpensive to check, and if there
406 	 * is anything runnable we won't have to wait for the poke.
407 	 */
408 	if (cpup->cpu_disp->disp_nrunnable != 0) {
409 		if (hset_update) {
410 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
411 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
412 		}
413 		sti();
414 		return;
415 	}
416 
417 	mach_cpu_idle();
418 
419 	/*
420 	 * We're no longer halted
421 	 */
422 	if (hset_update) {
423 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
424 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
425 	}
426 }
427 
428 
429 /*
430  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
431  * Otherwise, see if other CPUs in the cpu partition are halted and need to
432  * be woken up so that they can steal the thread we placed on this CPU.
433  * This function is only used on MP systems.
434  */
435 static void
436 cpu_wakeup(cpu_t *cpu, int bound)
437 {
438 	uint_t		cpu_found;
439 	int		result;
440 	cpupart_t	*cp;
441 
442 	cp = cpu->cpu_part;
443 	if (CPU_IN_SET(cp->cp_mach->mc_haltset, cpu->cpu_id)) {
444 		/*
445 		 * Clear the halted bit for that CPU since it will be
446 		 * poked in a moment.
447 		 */
448 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpu->cpu_id);
449 		/*
450 		 * We may find the current CPU present in the halted cpuset
451 		 * if we're in the context of an interrupt that occurred
452 		 * before we had a chance to clear our bit in cpu_idle().
453 		 * Poking ourself is obviously unnecessary, since if
454 		 * we're here, we're not halted.
455 		 */
456 		if (cpu != CPU)
457 			poke_cpu(cpu->cpu_id);
458 		return;
459 	} else {
460 		/*
461 		 * This cpu isn't halted, but it's idle or undergoing a
462 		 * context switch. No need to awaken anyone else.
463 		 */
464 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
465 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
466 			return;
467 	}
468 
469 	/*
470 	 * No need to wake up other CPUs if the thread we just enqueued
471 	 * is bound.
472 	 */
473 	if (bound)
474 		return;
475 
476 
477 	/*
478 	 * See if there's any other halted CPUs. If there are, then
479 	 * select one, and awaken it.
480 	 * It's possible that after we find a CPU, somebody else
481 	 * will awaken it before we get the chance.
482 	 * In that case, look again.
483 	 */
484 	do {
485 		CPUSET_FIND(cp->cp_mach->mc_haltset, cpu_found);
486 		if (cpu_found == CPUSET_NOTINSET)
487 			return;
488 
489 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
490 		CPUSET_ATOMIC_XDEL(cp->cp_mach->mc_haltset, cpu_found, result);
491 	} while (result < 0);
492 
493 	if (cpu_found != CPU->cpu_id)
494 		poke_cpu(cpu_found);
495 }
496 
497 #ifndef __xpv
498 /*
499  * Idle the present CPU until awoken via touching its monitored line
500  */
501 static void
502 cpu_idle_mwait(void)
503 {
504 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
505 	cpu_t			*cpup = CPU;
506 	processorid_t		cpun = cpup->cpu_id;
507 	cpupart_t		*cp = cpup->cpu_part;
508 	int			hset_update = 1;
509 
510 	/*
511 	 * Set our mcpu_mwait here, so we can tell if anyone trys to
512 	 * wake us between now and when we call mwait.  No other cpu will
513 	 * attempt to set our mcpu_mwait until we add ourself to the haltset.
514 	 */
515 	*mcpu_mwait = MWAIT_HALTED;
516 
517 	/*
518 	 * If this CPU is online, and there's multiple CPUs
519 	 * in the system, then we should notate our halting
520 	 * by adding ourselves to the partition's halted CPU
521 	 * bitmap. This allows other CPUs to find/awaken us when
522 	 * work becomes available.
523 	 */
524 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
525 		hset_update = 0;
526 
527 	/*
528 	 * Add ourselves to the partition's halted CPUs bitmask
529 	 * and set our HALTED flag, if necessary.
530 	 *
531 	 * When a thread becomes runnable, it is placed on the queue
532 	 * and then the halted cpuset is checked to determine who
533 	 * (if anyone) should be awoken. We therefore need to first
534 	 * add ourselves to the halted cpuset, and and then check if there
535 	 * is any work available.
536 	 *
537 	 * Note that memory barriers after updating the HALTED flag
538 	 * are not necessary since an atomic operation (updating the bitmap)
539 	 * immediately follows. On x86 the atomic operation acts as a
540 	 * memory barrier for the update of cpu_disp_flags.
541 	 */
542 	if (hset_update) {
543 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
544 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
545 	}
546 
547 	/*
548 	 * Check to make sure there's really nothing to do.
549 	 * Work destined for this CPU may become available after
550 	 * this check. We'll be notified through the clearing of our
551 	 * bit in the halted CPU bitmask, and a write to our mcpu_mwait.
552 	 *
553 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
554 	 */
555 	if (disp_anywork()) {
556 		if (hset_update) {
557 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
558 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
559 		}
560 		return;
561 	}
562 
563 	/*
564 	 * We're on our way to being halted.
565 	 * To avoid a lost wakeup, arm the monitor before checking if another
566 	 * cpu wrote to mcpu_mwait to wake us up.
567 	 */
568 	i86_monitor(mcpu_mwait, 0, 0);
569 	if (*mcpu_mwait == MWAIT_HALTED) {
570 		tlb_going_idle();
571 		i86_mwait(0, 0);
572 		tlb_service();
573 	}
574 
575 	/*
576 	 * We're no longer halted
577 	 */
578 	if (hset_update) {
579 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
580 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
581 	}
582 }
583 
584 /*
585  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
586  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
587  * need to be woken up so that they can steal the thread we placed on this CPU.
588  * This function is only used on MP systems.
589  */
590 static void
591 cpu_wakeup_mwait(cpu_t *cp, int bound)
592 {
593 	cpupart_t	*cpu_part;
594 	uint_t		cpu_found;
595 	int		result;
596 
597 	cpu_part = cp->cpu_part;
598 
599 	/*
600 	 * Clear the halted bit for that CPU since it will be woken up
601 	 * in a moment.
602 	 */
603 	if (CPU_IN_SET(cpu_part->cp_mach->mc_haltset, cp->cpu_id)) {
604 		/*
605 		 * Clear the halted bit for that CPU since it will be
606 		 * poked in a moment.
607 		 */
608 		CPUSET_ATOMIC_DEL(cpu_part->cp_mach->mc_haltset, cp->cpu_id);
609 		/*
610 		 * We may find the current CPU present in the halted cpuset
611 		 * if we're in the context of an interrupt that occurred
612 		 * before we had a chance to clear our bit in cpu_idle().
613 		 * Waking ourself is obviously unnecessary, since if
614 		 * we're here, we're not halted.
615 		 *
616 		 * monitor/mwait wakeup via writing to our cache line is
617 		 * harmless and less expensive than always checking if we
618 		 * are waking ourself which is an uncommon case.
619 		 */
620 		MWAIT_WAKEUP(cp);	/* write to monitored line */
621 		return;
622 	} else {
623 		/*
624 		 * This cpu isn't halted, but it's idle or undergoing a
625 		 * context switch. No need to awaken anyone else.
626 		 */
627 		if (cp->cpu_thread == cp->cpu_idle_thread ||
628 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
629 			return;
630 	}
631 
632 	/*
633 	 * No need to wake up other CPUs if the thread we just enqueued
634 	 * is bound.
635 	 */
636 	if (bound)
637 		return;
638 
639 
640 	/*
641 	 * See if there's any other halted CPUs. If there are, then
642 	 * select one, and awaken it.
643 	 * It's possible that after we find a CPU, somebody else
644 	 * will awaken it before we get the chance.
645 	 * In that case, look again.
646 	 */
647 	do {
648 		CPUSET_FIND(cpu_part->cp_mach->mc_haltset, cpu_found);
649 		if (cpu_found == CPUSET_NOTINSET)
650 			return;
651 
652 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
653 		CPUSET_ATOMIC_XDEL(cpu_part->cp_mach->mc_haltset, cpu_found,
654 		    result);
655 	} while (result < 0);
656 
657 	/*
658 	 * Do not check if cpu_found is ourself as monitor/mwait wakeup is
659 	 * cheap.
660 	 */
661 	MWAIT_WAKEUP(cpu[cpu_found]);	/* write to monitored line */
662 }
663 #endif
664 
665 void (*cpu_pause_handler)(volatile char *) = NULL;
666 
667 static int
668 mp_disable_intr(int cpun)
669 {
670 	/*
671 	 * switch to the offline cpu
672 	 */
673 	affinity_set(cpun);
674 	/*
675 	 * raise ipl to just below cross call
676 	 */
677 	splx(XC_MED_PIL-1);
678 	/*
679 	 *	set base spl to prevent the next swtch to idle from
680 	 *	lowering back to ipl 0
681 	 */
682 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
683 	set_base_spl();
684 	affinity_clear();
685 	return (DDI_SUCCESS);
686 }
687 
688 static void
689 mp_enable_intr(int cpun)
690 {
691 	/*
692 	 * switch to the online cpu
693 	 */
694 	affinity_set(cpun);
695 	/*
696 	 * clear the interrupt active mask
697 	 */
698 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
699 	set_base_spl();
700 	(void) spl0();
701 	affinity_clear();
702 }
703 
704 static void
705 mach_get_platform(int owner)
706 {
707 	void		**srv_opsp;
708 	void		**clt_opsp;
709 	int		i;
710 	int		total_ops;
711 
712 	/* fix up psm ops */
713 	srv_opsp = (void **)mach_set[0];
714 	clt_opsp = (void **)mach_set[owner];
715 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
716 		total_ops = sizeof (struct psm_ops_ver01) /
717 		    sizeof (void (*)(void));
718 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
719 		/* no psm_notify_func */
720 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
721 		    sizeof (void (*)(void));
722 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
723 		/* no psm_timer funcs */
724 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
725 		    sizeof (void (*)(void));
726 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
727 		/* no psm_preshutdown function */
728 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
729 		    sizeof (void (*)(void));
730 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
731 		/* no psm_preshutdown function */
732 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
733 		    sizeof (void (*)(void));
734 	else
735 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
736 
737 	/*
738 	 * Save the version of the PSM module, in case we need to
739 	 * bahave differently based on version.
740 	 */
741 	mach_ver[0] = mach_ver[owner];
742 
743 	for (i = 0; i < total_ops; i++)
744 		if (clt_opsp[i] != NULL)
745 			srv_opsp[i] = clt_opsp[i];
746 }
747 
748 static void
749 mach_construct_info()
750 {
751 	struct psm_sw *swp;
752 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
753 	int	conflict_owner = 0;
754 
755 	if (psmsw->psw_forw == psmsw)
756 		panic("No valid PSM modules found");
757 	mutex_enter(&psmsw_lock);
758 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
759 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
760 			continue;
761 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
762 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
763 		mach_cnt[swp->psw_infop->p_owner]++;
764 	}
765 	mutex_exit(&psmsw_lock);
766 
767 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
768 
769 	/* check to see are there any conflicts */
770 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
771 		conflict_owner = PSM_OWN_EXCLUSIVE;
772 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
773 		conflict_owner = PSM_OWN_OVERRIDE;
774 	if (conflict_owner) {
775 		/* remove all psm modules except uppc */
776 		cmn_err(CE_WARN,
777 		    "Conflicts detected on the following PSM modules:");
778 		mutex_enter(&psmsw_lock);
779 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
780 			if (swp->psw_infop->p_owner == conflict_owner)
781 				cmn_err(CE_WARN, "%s ",
782 				    swp->psw_infop->p_mach_idstring);
783 		}
784 		mutex_exit(&psmsw_lock);
785 		cmn_err(CE_WARN,
786 		    "Setting the system back to SINGLE processor mode!");
787 		cmn_err(CE_WARN,
788 		    "Please edit /etc/mach to remove the invalid PSM module.");
789 		return;
790 	}
791 
792 	if (mach_set[PSM_OWN_EXCLUSIVE])
793 		mach_get_platform(PSM_OWN_EXCLUSIVE);
794 
795 	if (mach_set[PSM_OWN_OVERRIDE])
796 		mach_get_platform(PSM_OWN_OVERRIDE);
797 }
798 
799 static void
800 mach_init()
801 {
802 	struct psm_ops  *pops;
803 
804 	mach_construct_info();
805 
806 	pops = mach_set[0];
807 
808 	/* register the interrupt and clock initialization rotuines */
809 	picinitf = mach_picinit;
810 	clkinitf = mach_clkinit;
811 	psm_get_clockirq = pops->psm_get_clockirq;
812 
813 	/* register the interrupt setup code */
814 	slvltovect = mach_softlvl_to_vect;
815 	addspl	= pops->psm_addspl;
816 	delspl	= pops->psm_delspl;
817 
818 	if (pops->psm_translate_irq)
819 		psm_translate_irq = pops->psm_translate_irq;
820 	if (pops->psm_intr_ops)
821 		psm_intr_ops = pops->psm_intr_ops;
822 
823 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
824 	/*
825 	 * Time-of-day functionality now handled in TOD modules.
826 	 * (Warn about PSM modules that think that we're going to use
827 	 * their ops vectors.)
828 	 */
829 	if (pops->psm_tod_get)
830 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
831 		    (void *)pops->psm_tod_get);
832 
833 	if (pops->psm_tod_set)
834 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
835 		    (void *)pops->psm_tod_set);
836 #endif
837 
838 	if (pops->psm_notify_error) {
839 		psm_notify_error = mach_notify_error;
840 		notify_error = pops->psm_notify_error;
841 	}
842 
843 	(*pops->psm_softinit)();
844 
845 	/*
846 	 * Initialize the dispatcher's function hooks
847 	 * to enable CPU halting when idle.
848 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
849 	 * or idle_cpu_prefer_mwait is not set.
850 	 * Allocate monitor/mwait buffer for cpu0.
851 	 */
852 	if (idle_cpu_use_hlt) {
853 		idle_cpu = cpu_idle;
854 #ifndef __xpv
855 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
856 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
857 			/*
858 			 * Protect ourself from insane mwait size.
859 			 */
860 			if (CPU->cpu_m.mcpu_mwait == NULL) {
861 #ifdef DEBUG
862 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
863 				    "handle cpu 0 mwait size.");
864 #endif
865 				idle_cpu_prefer_mwait = 0;
866 				idle_cpu = cpu_idle;
867 			} else {
868 				idle_cpu = cpu_idle_mwait;
869 			}
870 		} else {
871 			idle_cpu = cpu_idle;
872 		}
873 #endif
874 	}
875 
876 	mach_smpinit();
877 }
878 
879 static void
880 mach_smpinit(void)
881 {
882 	struct psm_ops  *pops;
883 	processorid_t cpu_id;
884 	int cnt;
885 	cpuset_t cpumask;
886 
887 	pops = mach_set[0];
888 
889 	cpu_id = -1;
890 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
891 	for (cnt = 0, CPUSET_ZERO(cpumask); cpu_id != -1; cnt++) {
892 		CPUSET_ADD(cpumask, cpu_id);
893 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
894 	}
895 
896 	mp_cpus = cpumask;
897 
898 	/* MP related routines */
899 	ap_mlsetup = pops->psm_post_cpu_start;
900 	send_dirintf = pops->psm_send_ipi;
901 
902 	/* optional MP related routines */
903 	if (pops->psm_shutdown)
904 		psm_shutdownf = pops->psm_shutdown;
905 	if (pops->psm_preshutdown)
906 		psm_preshutdownf = pops->psm_preshutdown;
907 	if (pops->psm_notify_func)
908 		psm_notifyf = pops->psm_notify_func;
909 	if (pops->psm_set_idlecpu)
910 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
911 	if (pops->psm_unset_idlecpu)
912 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
913 
914 	psm_clkinit = pops->psm_clkinit;
915 
916 	if (pops->psm_timer_reprogram)
917 		psm_timer_reprogram = pops->psm_timer_reprogram;
918 
919 	if (pops->psm_timer_enable)
920 		psm_timer_enable = pops->psm_timer_enable;
921 
922 	if (pops->psm_timer_disable)
923 		psm_timer_disable = pops->psm_timer_disable;
924 
925 	if (pops->psm_post_cyclic_setup)
926 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
927 
928 	if (pops->psm_state)
929 		psm_state = pops->psm_state;
930 
931 	/* check for multiple cpu's */
932 	if (cnt < 2)
933 		return;
934 
935 	/* check for MP platforms */
936 	if (pops->psm_cpu_start == NULL)
937 		return;
938 
939 	/*
940 	 * Set the dispatcher hook to enable cpu "wake up"
941 	 * when a thread becomes runnable.
942 	 */
943 	if (idle_cpu_use_hlt) {
944 		disp_enq_thread = cpu_wakeup;
945 #ifndef __xpv
946 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
947 			disp_enq_thread = cpu_wakeup_mwait;
948 #endif
949 	}
950 
951 	if (pops->psm_disable_intr)
952 		psm_disable_intr = pops->psm_disable_intr;
953 	if (pops->psm_enable_intr)
954 		psm_enable_intr  = pops->psm_enable_intr;
955 
956 	psm_get_ipivect = pops->psm_get_ipivect;
957 
958 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
959 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
960 	    (caddr_t)X_CALL_HIPRI, NULL, NULL, NULL);
961 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
962 	    (*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
963 	    (caddr_t)X_CALL_MEDPRI, NULL, NULL, NULL);
964 
965 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
966 }
967 
968 static void
969 mach_picinit()
970 {
971 	struct psm_ops  *pops;
972 
973 	pops = mach_set[0];
974 
975 	/* register the interrupt handlers */
976 	setlvl = pops->psm_intr_enter;
977 	setlvlx = pops->psm_intr_exit;
978 
979 	/* initialize the interrupt hardware */
980 	(*pops->psm_picinit)();
981 
982 	/* set interrupt mask for current ipl */
983 	setspl = pops->psm_setspl;
984 	cli();
985 	setspl(CPU->cpu_pri);
986 }
987 
988 uint_t	cpu_freq;	/* MHz */
989 uint64_t cpu_freq_hz;	/* measured (in hertz) */
990 
991 #define	MEGA_HZ		1000000
992 
993 #ifdef __xpv
994 
995 int xpv_cpufreq_workaround = 1;
996 int xpv_cpufreq_verbose = 0;
997 
998 #else	/* __xpv */
999 
1000 static uint64_t
1001 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1002 {
1003 	uint64_t cpu_hz;
1004 
1005 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1006 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1007 		return (0);
1008 
1009 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1010 
1011 	return (cpu_hz);
1012 }
1013 
1014 #endif	/* __xpv */
1015 
1016 static uint64_t
1017 mach_getcpufreq(void)
1018 {
1019 #if defined(__xpv)
1020 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1021 	uint64_t cpu_hz;
1022 
1023 	/*
1024 	 * During dom0 bringup, it was noted that on at least one older
1025 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1026 	 * value that is quite wrong (the 3.06GHz clock was reported
1027 	 * as 4.77GHz)
1028 	 *
1029 	 * The curious thing is, that if you stop the kernel at entry,
1030 	 * breakpoint here and inspect the value with kmdb, the value
1031 	 * is correct - but if you don't stop and simply enable the
1032 	 * printf statement (below), you can see the bad value printed
1033 	 * here.  Almost as if something kmdb did caused the hypervisor to
1034 	 * figure it out correctly.  And, note that the hypervisor
1035 	 * eventually -does- figure it out correctly ... if you look at
1036 	 * the field later in the life of dom0, it is correct.
1037 	 *
1038 	 * For now, on dom0, we employ a slightly cheesy workaround of
1039 	 * using the DOM0_PHYSINFO hypercall.
1040 	 */
1041 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1042 		xen_sysctl_t op0, *op = &op0;
1043 
1044 		op->cmd = XEN_SYSCTL_physinfo;
1045 		op->interface_version = XEN_SYSCTL_INTERFACE_VERSION;
1046 		if (HYPERVISOR_sysctl(op) != 0)
1047 			panic("physinfo op refused");
1048 
1049 		cpu_hz = 1000 * (uint64_t)op->u.physinfo.cpu_khz;
1050 	} else {
1051 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1052 
1053 		if (vti->tsc_shift < 0)
1054 			cpu_hz <<= -vti->tsc_shift;
1055 		else
1056 			cpu_hz >>= vti->tsc_shift;
1057 	}
1058 
1059 	if (xpv_cpufreq_verbose)
1060 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1061 		    "cpu_hz %" PRId64 "Hz\n",
1062 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1063 
1064 	return (cpu_hz);
1065 #else	/* __xpv */
1066 	uint32_t pit_counter;
1067 	uint64_t processor_clks;
1068 
1069 	if (x86_feature & X86_TSC) {
1070 		/*
1071 		 * We have a TSC. freq_tsc() knows how to measure the number
1072 		 * of clock cycles sampled against the PIT.
1073 		 */
1074 		ulong_t flags = clear_int_flag();
1075 		processor_clks = freq_tsc(&pit_counter);
1076 		restore_int_flag(flags);
1077 		return (mach_calchz(pit_counter, &processor_clks));
1078 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1079 #if defined(__amd64)
1080 		panic("mach_getcpufreq: no TSC!");
1081 #elif defined(__i386)
1082 		/*
1083 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1084 		 * for which freq_notsc() knows how to measure the number of
1085 		 * elapsed clock cycles sampled against the PIT
1086 		 */
1087 		ulong_t flags = clear_int_flag();
1088 		processor_clks = freq_notsc(&pit_counter);
1089 		restore_int_flag(flags);
1090 		return (mach_calchz(pit_counter, &processor_clks));
1091 #endif	/* __i386 */
1092 	}
1093 
1094 	/* We do not know how to calculate cpu frequency for this cpu. */
1095 	return (0);
1096 #endif	/* __xpv */
1097 }
1098 
1099 /*
1100  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1101  * to this array, instead improve the accuracy of the algorithm that determines
1102  * the clock speed of the processor or extend the implementation to support the
1103  * vendor as appropriate. This is here only to support adjusting the speed on
1104  * older slower processors that mach_fixcpufreq() would not be able to account
1105  * for otherwise.
1106  */
1107 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1108 
1109 /*
1110  * On fast processors the clock frequency that is measured may be off by
1111  * a few MHz from the value printed on the part. This is a combination of
1112  * the factors that for such fast parts being off by this much is within
1113  * the tolerances for manufacture and because of the difficulties in the
1114  * measurement that can lead to small error. This function uses some
1115  * heuristics in order to tweak the value that was measured to match what
1116  * is most likely printed on the part.
1117  *
1118  * Some examples:
1119  * 	AMD Athlon 1000 mhz measured as 998 mhz
1120  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1121  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1122  *
1123  * If in the future this function is no longer sufficient to correct
1124  * for the error in the measurement, then the algorithm used to perform
1125  * the measurement will have to be improved in order to increase accuracy
1126  * rather than adding horrible and questionable kludges here.
1127  *
1128  * This is called after the cyclics subsystem because of the potential
1129  * that the heuristics within may give a worse estimate of the clock
1130  * frequency than the value that was measured.
1131  */
1132 static void
1133 mach_fixcpufreq(void)
1134 {
1135 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1136 
1137 	freq = (uint32_t)cpu_freq;
1138 
1139 	/*
1140 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1141 	 * measured speed taking into account that the 667 MHz parts were
1142 	 * the first to round-up.
1143 	 */
1144 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1145 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1146 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1147 
1148 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1149 	mul = (freq + 25) / 50;
1150 	near50 = mul * 50;
1151 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1152 
1153 	/* Find the closer of the two */
1154 	if (delta66 < delta50) {
1155 		fixed = near66;
1156 		delta = delta66;
1157 	} else {
1158 		fixed = near50;
1159 		delta = delta50;
1160 	}
1161 
1162 	if (fixed > INT_MAX)
1163 		return;
1164 
1165 	/*
1166 	 * Some older parts have a core clock frequency that is not an
1167 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1168 	 * clock frequencies is closer to the measured value than any
1169 	 * of the integral multiples of 50 an 66, and if so set fixed
1170 	 * and delta appropriately to represent the closest value.
1171 	 */
1172 	i = sizeof (x86_cpu_freq) / sizeof (int);
1173 	while (i > 0) {
1174 		i--;
1175 
1176 		if (x86_cpu_freq[i] <= freq) {
1177 			mul = freq - x86_cpu_freq[i];
1178 
1179 			if (mul < delta) {
1180 				fixed = x86_cpu_freq[i];
1181 				delta = mul;
1182 			}
1183 
1184 			break;
1185 		}
1186 
1187 		mul = x86_cpu_freq[i] - freq;
1188 
1189 		if (mul < delta) {
1190 			fixed = x86_cpu_freq[i];
1191 			delta = mul;
1192 		}
1193 	}
1194 
1195 	/*
1196 	 * Set a reasonable maximum for how much to correct the measured
1197 	 * result by. This check is here to prevent the adjustment made
1198 	 * by this function from being more harm than good. It is entirely
1199 	 * possible that in the future parts will be made that are not
1200 	 * integral multiples of 66 or 50 in clock frequency or that
1201 	 * someone may overclock a part to some odd frequency. If the
1202 	 * measured value is farther from the corrected value than
1203 	 * allowed, then assume the corrected value is in error and use
1204 	 * the measured value.
1205 	 */
1206 	if (6 < delta)
1207 		return;
1208 
1209 	cpu_freq = (int)fixed;
1210 }
1211 
1212 
1213 static int
1214 machhztomhz(uint64_t cpu_freq_hz)
1215 {
1216 	uint64_t cpu_mhz;
1217 
1218 	/* Round to nearest MHZ */
1219 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1220 
1221 	if (cpu_mhz > INT_MAX)
1222 		return (0);
1223 
1224 	return ((int)cpu_mhz);
1225 
1226 }
1227 
1228 
1229 static int
1230 mach_clkinit(int preferred_mode, int *set_mode)
1231 {
1232 	struct psm_ops  *pops;
1233 	int resolution;
1234 
1235 	pops = mach_set[0];
1236 
1237 	cpu_freq_hz = mach_getcpufreq();
1238 
1239 	cpu_freq = machhztomhz(cpu_freq_hz);
1240 
1241 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1242 		tsc_gethrtime_enable = 0;
1243 
1244 #ifndef __xpv
1245 	if (tsc_gethrtime_enable) {
1246 		tsc_hrtimeinit(cpu_freq_hz);
1247 	} else
1248 #endif
1249 	{
1250 		if (pops->psm_hrtimeinit)
1251 			(*pops->psm_hrtimeinit)();
1252 		gethrtimef = pops->psm_gethrtime;
1253 		gethrtimeunscaledf = gethrtimef;
1254 		/* scalehrtimef will remain dummy */
1255 	}
1256 
1257 	mach_fixcpufreq();
1258 
1259 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1260 		if (preferred_mode == TIMER_ONESHOT) {
1261 
1262 			resolution = (*pops->psm_clkinit)(0);
1263 			if (resolution != 0)  {
1264 				*set_mode = TIMER_ONESHOT;
1265 				return (resolution);
1266 			}
1267 		}
1268 
1269 		/*
1270 		 * either periodic mode was requested or could not set to
1271 		 * one-shot mode
1272 		 */
1273 		resolution = (*pops->psm_clkinit)(hz);
1274 		/*
1275 		 * psm should be able to do periodic, so we do not check
1276 		 * for return value of psm_clkinit here.
1277 		 */
1278 		*set_mode = TIMER_PERIODIC;
1279 		return (resolution);
1280 	} else {
1281 		/*
1282 		 * PSMI interface prior to PSMI_3 does not define a return
1283 		 * value for psm_clkinit, so the return value is ignored.
1284 		 */
1285 		(void) (*pops->psm_clkinit)(hz);
1286 		*set_mode = TIMER_PERIODIC;
1287 		return (nsec_per_tick);
1288 	}
1289 }
1290 
1291 
1292 /*ARGSUSED*/
1293 static int
1294 mach_softlvl_to_vect(int ipl)
1295 {
1296 	setsoftint = av_set_softint_pending;
1297 	kdisetsoftint = kdi_av_set_softint_pending;
1298 
1299 	return (PSM_SV_SOFTWARE);
1300 }
1301 
1302 #ifdef DEBUG
1303 /*
1304  * This is here to allow us to simulate cpus that refuse to start.
1305  */
1306 cpuset_t cpufailset;
1307 #endif
1308 
1309 int
1310 mach_cpu_start(struct cpu *cp, void *ctx)
1311 {
1312 	struct psm_ops *pops = mach_set[0];
1313 	processorid_t id = cp->cpu_id;
1314 
1315 #ifdef DEBUG
1316 	if (CPU_IN_SET(cpufailset, id))
1317 		return (0);
1318 #endif
1319 	return ((*pops->psm_cpu_start)(id, ctx));
1320 }
1321 
1322 int
1323 mach_cpuid_start(processorid_t id, void *ctx)
1324 {
1325 	struct psm_ops *pops = mach_set[0];
1326 
1327 #ifdef DEBUG
1328 	if (CPU_IN_SET(cpufailset, id))
1329 		return (0);
1330 #endif
1331 	return ((*pops->psm_cpu_start)(id, ctx));
1332 }
1333 
1334 /*ARGSUSED*/
1335 static int
1336 mach_translate_irq(dev_info_t *dip, int irqno)
1337 {
1338 	return (irqno);	/* default to NO translation */
1339 }
1340 
1341 static void
1342 mach_notify_error(int level, char *errmsg)
1343 {
1344 	/*
1345 	 * SL_FATAL is pass in once panicstr is set, deliver it
1346 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1347 	 * codes for the psmi handler
1348 	 */
1349 	if (level & SL_FATAL)
1350 		(*notify_error)(CE_PANIC, errmsg);
1351 	else if (level & SL_WARN)
1352 		(*notify_error)(CE_WARN, errmsg);
1353 	else if (level & SL_NOTE)
1354 		(*notify_error)(CE_NOTE, errmsg);
1355 	else if (level & SL_CONSOLE)
1356 		(*notify_error)(CE_CONT, errmsg);
1357 }
1358 
1359 /*
1360  * It provides the default basic intr_ops interface for the new DDI
1361  * interrupt framework if the PSM doesn't have one.
1362  *
1363  * Input:
1364  * dip     - pointer to the dev_info structure of the requested device
1365  * hdlp    - pointer to the internal interrupt handle structure for the
1366  *	     requested interrupt
1367  * intr_op - opcode for this call
1368  * result  - pointer to the integer that will hold the result to be
1369  *	     passed back if return value is PSM_SUCCESS
1370  *
1371  * Output:
1372  * return value is either PSM_SUCCESS or PSM_FAILURE
1373  */
1374 static int
1375 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1376     psm_intr_op_t intr_op, int *result)
1377 {
1378 	struct intrspec *ispec;
1379 
1380 	switch (intr_op) {
1381 	case PSM_INTR_OP_CHECK_MSI:
1382 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1383 		    DDI_INTR_TYPE_MSIX);
1384 		break;
1385 	case PSM_INTR_OP_ALLOC_VECTORS:
1386 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1387 			*result = 1;
1388 		else
1389 			*result = 0;
1390 		break;
1391 	case PSM_INTR_OP_FREE_VECTORS:
1392 		break;
1393 	case PSM_INTR_OP_NAVAIL_VECTORS:
1394 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1395 			*result = 1;
1396 		else
1397 			*result = 0;
1398 		break;
1399 	case PSM_INTR_OP_XLATE_VECTOR:
1400 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1401 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1402 		break;
1403 	case PSM_INTR_OP_GET_CAP:
1404 		*result = 0;
1405 		break;
1406 	case PSM_INTR_OP_GET_PENDING:
1407 	case PSM_INTR_OP_CLEAR_MASK:
1408 	case PSM_INTR_OP_SET_MASK:
1409 	case PSM_INTR_OP_GET_SHARED:
1410 	case PSM_INTR_OP_SET_PRI:
1411 	case PSM_INTR_OP_SET_CAP:
1412 	case PSM_INTR_OP_SET_CPU:
1413 	case PSM_INTR_OP_GET_INTR:
1414 	default:
1415 		return (PSM_FAILURE);
1416 	}
1417 	return (PSM_SUCCESS);
1418 }
1419 /*
1420  * Return 1 if CMT load balancing policies should be
1421  * implemented across instances of the specified hardware
1422  * sharing relationship.
1423  */
1424 int
1425 pg_cmt_load_bal_hw(pghw_type_t hw)
1426 {
1427 	if (hw == PGHW_IPIPE ||
1428 	    hw == PGHW_FPU ||
1429 	    hw == PGHW_CHIP)
1430 		return (1);
1431 	else
1432 		return (0);
1433 }
1434 /*
1435  * Return 1 if thread affinity polices should be implemented
1436  * for instances of the specifed hardware sharing relationship.
1437  */
1438 int
1439 pg_cmt_affinity_hw(pghw_type_t hw)
1440 {
1441 	if (hw == PGHW_CACHE)
1442 		return (1);
1443 	else
1444 		return (0);
1445 }
1446